| Benchmark | DeepSeek-V3 | Gemini 2.5 Pro |
|---|---|---|
| AA Agentic Index | 1.6 | 7.2 |
| AA Intelligence | 15.4 | 27 |
| AA-LCR | 41.3 | 66 |
| AA-Omniscience | -37.6 | -14.3 |
| aider_polyglot | 55.1 | 82.2 |
| AIME 2025 | 51.3 | 88.3 |
| AIR-Bench 2024 | 40.8 | 73.6 |
| anthropic_red_team | 97.1 | 99.5 |
| Artificial Analysis Coding Index | 23 | 46.7 |
| bbq | 96.7 | 96.4 |
| critpt | 0 | 2.6 |
| gdpval | 0 | 8.5 |
| GPQA Diamond | 68.4 | 84.4 |
| harmbench | 49.7 | 65.4 |
| HLE | 5.2 | 22.5 |
| HMMT 2025 | 27.5 | 65.7 |
| hmmt_feb_2025 | 29.2 | 82.5 |
| IFBench | 41 | 49 |
| LiveCodeBench | 49.6 | 82.7 |
| livecodebench_easy | 83.3 | 98.8 |
| livecodebench_hard | 14.2 | 59.4 |
| livecodebench_medium | 53.5 | 90.6 |
| MMLU-Pro | 81.2 | 86 |
| multichallenge | 31.4 | 53.6 |
| OmniScience Accuracy | 25.4 | 39 |
| OmniScience Non-Hallucination | 23.3 | 12.6 |
| scicode | 35.8 | 43 |
| simple_safety_tests | 95.3 | 97 |
| simplebench | 27.2 | 62.4 |
| simpleqa | 27.7 | 50.8 |
| SWE-bench Verified | 42 | 67.2 |
| TauBench V3 - Banking | 4.7 | 9.7 |
| Terminal-Bench 2.1 | 16.9 | 28.5 |
| Terminal-Bench Hard | 15.2 | 26.5 |
| vectara_answer_rate | 97.5 | 99.1 |
| vectara_avg_summary_length | 81.7 | 106.4 |
| vectara_factual_consistency | 93.9 | 93 |
| vectara_hallucination_rate ↓ | 6.1 | 7 |
| xstest | 97.1 | 98.7 |
| τ²-Bench | 32.5 | 54 |
| τ²-Bench Telecom (AA run) | 47.1 | 54.1 |
| τ³-Bench | 4.7 | 9.3 |
Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.